A common statistical misunderstanding in Psychology and Neuroscience: Do we need normally distributed independent or dependent variables for linear regression to work?

Our ability to draw conclusions from experiments often relies on the application of inferential statistics, such as those included within the general-linear-modeling framework. The purpose of this commentary is to direct attention a common fallacy that can be observed in paper reviews and professional exchanges in the field of Psychology and Neuroscience pertaining to multiple regression analysis. The fallacy concerns the confusion of the requirements on the residuals of a linear-regression model with requirements on the dependent or independent variables themselves. A simple thought experiment, algebraic demonstration, and simulation study can demonstrate this misconception. It is our hope that consideration of the fallacy may help clarify distinctions between technical statistical issues and conceptual interpretation of research findings.